The integration of large language models such as ChatGPT has raised concerns about stylistic homogenization in scholarly writing. While scientific literature shows clear LLM-driven shifts, e.g., increased lexical markers and reduced cohesion (Bao et al., 2025; Kousha & Thelwall, 2024), this study examines whether similar changes appear in humanities thesis and dissertation titles. Drawing on 8,631 unique MA and PhD titles from ProQuest in History, Religion, Literature, Philosophy, and Musicology, linguistic features were compared between 2015 (pre-AI) and 2025 (post-AI stabilization). Five dimensions were analyzed: word length, informativity, lexical diversity, syntactic structure, and semantic content. Results reveal remarkable stability across most metrics (title length ~12–13 words, informativity ~67%, lexical diversity near 100%). Only a modest increase in compound structures (70% to 74%) occurred, reflecting amplification of existing humanities conventions rather than disruption. The brevity of titles and extended human supervision appear to limit deep LLM intervention. These findings contrast with scientific fields and highlight the resilience of disciplinary norms in graduate scholarship.